Second-order optimisation strategies for neural network quantum states

Abstract

The Variational Monte Carlo method has recently seen important advances through the use of neural network quantum states. While more and more sophisticated ansätze have been designed to tackle a wide variety of quantum many-body problems, modest progress has been made on the associated optimisation algorithms. In this work, we revisit the Kronecker-Factored Approximate Curvature, an optimiser that has been used extensively in a variety of simulations. We suggest improvements on the scaling and the direction of this optimiser, and find that they substantially increase its performance at a negligible additional cost. We also reformulate the Variational Monte Carlo approach in a game theory framework, to propose a novel optimiser based on decision geometry. We find that, on a practical test case for continuous systems, this new optimiser consistently outperforms any of the KFAC improvements in terms of stability, accuracy and speed of convergence. Beyond Variational Monte Carlo, the versatility of this approach suggests that decision geometry could provide a solid foundation for accelerating a broad class of machine learning algorithms.